Performance comparison of classical neural network algorithms in emotion recognition

Emotion recognition with computer-assisted methods has been an important tool for psychological analysis. Traditional approaches based on designing and extracting various features of multi-channel neurophysiological signals may require extensive knowledge of this field. This can be a sophisticated task for any expert. In this paper, we examined a hybrid model that utilizes ‘Convolutional Neural Network (CNN)’ for feature extraction and ‘Recurrent Neural Network (RNN)’ that incorporates contextual information. Through the method proposed in this paper, the information that needs to be processed is optimized through multi-channel processing. At the same time, the signal data are packaged into different grid frames by the method of scale transformation, and finally the preprocessing is realized. The whole experiment process is carried out using the publicly available DEAP dataset. The final comparison result based on Valence and Arousal classification accuracy demonstrates that our method is effective.